How to set up message queues for asynchronous sending

Publishing messages to a message broker is a very fast operation, and we can leverage this to speed up our web services. We can delegate some tasks to background processes by making our web service publish a message to a queue instead. We can then have background consumer processes consume the messages and perform the delegated tasks as needed.

In this guide, we’ll be using RabbitMQ as a message broker. We’ll integrate RabbitMQ into a sample Flask application to defer sending mail to a background process. The application, Flaskr, may look familiar if you’ve previously completed the Flask Tutorial.

Let’s take a look at our signup view function:

Currently, the view function checks the database to make sure an account with that email address doesn’t already exist and then adds the new user to the database. The view then immediately sends a welcome email message to the user before finishing the request.

There are a couple of issues with this view that we can improve with a message queue. We don’t need to make the user wait for a response while we send the welcome email. It would be better if we can respond to the user as soon as possible. We also want a way to retry sending the welcome email if for some reason we’re unable to do it on the first try.

First, let’s run an instance of RabbitMQ in a Docker container:

Now let’s add some code to initialize a queue when our application starts:

We’re declaring a queue called “welcome_queue” that we’ll use to send messages to workers with the email address where the welcome email needs to be sent.

Next, we can update our signup view to publish a message to the queue instead of sending the welcome email.

Now we can write our worker code. The cool thing about using a message queue is that the worker process can run anywhere. It can run on dedicated worker servers, or maybe alongside your web service. This is what a simple worker script would look like:

We’re almost done! We’ve successfully decoupled sending our welcome email from the web service. Now the only thing missing is for us to retry sending the welcome email in case it fails to send for some reason. One clever way to add retry logic to your application is to use a dead letter queue.

We’ll add a second queue, the “retry_queue”, which we’ll use to temporarily place a message if we ever encounter an email send error. The messages in the retry queue will have an expiration date some time in the future. We’ll configure RabbitMQ such that when a message expires it will be placed back in our welcome queue ready to be picked up by a worker to attempt to send the email again:

And that’s it! We’ve successfully introduced asynchronous sending into our application using RabbitMQ. To download the full working example source code repository, check out my github repo.

Get more guides like this by subscribing to the blog. You’ll get an update each week with the latest posts from the Mailgun team. And if you’re not yet using Mailgun to send, receive, and track your application’s emails, you should sign up below!